Ilya Lesokhin schrieb:
Please don't forget to CC the lists.
I think you are wrong.
the following in avr.md is wrong:
;; "mulhisi3"
;; "umulhisi3"
(define_expand "<extend_u>mulhisi3"
[(parallel [(set (match_operand:SI 0 "register_operand" "")
(mult:SI (any_extend:SI (match_operand:HI 1
"register_operand" ""))
(any_extend:SI (match_operand:HI 2
"register_operand" ""))))
(clobber (reg:HI 26))
(clobber (reg:DI 18))])]
"AVR_HAVE_MUL"
"")
This is code from 4.7. The problem above is reported against 4.5.
you cant multiply two signed numbers this way, for example: -1*-1=1
and in 32bit its correct: 0xffffffff*0xffffffff = 1 (mod 2^32)
but using 16bit multiplication you get: 0xffff*0xffff=FFFE0001 (mod
2^32) != -1.
In a n = n*n multiplication, and if flags don't matter, it does not
matter if you take the numbers as signed or unsigned; just like that
does not matter for addition or subtraction.
Anyway, in the pattern you see a widening multiplication, not a n = n*n
vanilla one. The multiply is performed with 32 bits so that you get
0xffff * 0x0001 = 0xffffffff for example.
Johann
_______________________________________________
AVR-GCC-list mailing list
AVR-GCC-list@nongnu.org
https://lists.nongnu.org/mailman/listinfo/avr-gcc-list
